#setwd('/afs/inf.ed.ac.uk/user/s17/s1725186/Documents/PhD-Models/FirstPUModel/RMarkdowns')

library(tidyverse) ; library(reshape2) ; library(glue) ; library(plotly) ; library(dendextend)
library(RColorBrewer) ; library(viridis) ; require(gridExtra) ; library(GGally) ; library(ggpubr)
library(expss)
library(polycor)
library(foreach) ; library(doParallel)
library(knitr) ; library(kableExtra)
library(biomaRt)
library(clusterProfiler) ; library(ReactomePA) ; library(DOSE) ; library(org.Hs.eg.db)
library(WGCNA)

SFARI_colour_hue = function(r) {
  pal = c('#FF7631','#FFB100','#E8E328','#8CC83F','#62CCA6','#59B9C9','#b3b3b3','#808080','gray','#d9d9d9')[r]
}

Load preprocessed dataset (preprocessing code in 01_data_preprocessing.Rmd) and clustering (pipeline in 05_WGCNA.Rmd)

# Gandal dataset
load('./../Data/preprocessed_data.RData')
datExpr = datExpr %>% data.frame
rownames(datExpr) = datGenes$ensembl_gene_id
datMeta = datMeta %>% mutate(ID = paste0('X',description))


# GO Neuronal annotations: regex 'neuron' in GO functional annotations and label the genes that make a match as neuronal
GO_annotations = read.csv('./../Data/genes_GO_annotations.csv')
GO_neuronal = GO_annotations %>% filter(grepl('neuron', go_term)) %>% 
              mutate('ID'=as.character(ensembl_gene_id)) %>% 
              dplyr::select(-ensembl_gene_id) %>% distinct(ID) %>%
              mutate('Neuronal'=1)


# SFARI Genes
SFARI_genes = read_csv('./../../../SFARI/Data/SFARI_genes_01-03-2020_w_ensembl_IDs.csv')
SFARI_genes = SFARI_genes[!duplicated(SFARI_genes$ID) & !is.na(SFARI_genes$ID),]


# Clusterings
clusterings = read_csv('./../Data/clusters.csv')


# Update DE_info with SFARI and Neuronal information
genes_info = DE_info %>% mutate('ID'=datGenes$ensembl_gene_id, padj = adj.P.Val, log2FoldChange = logFC) %>% 
             left_join(SFARI_genes, by='ID') %>% 
             mutate(`gene-score`=ifelse(is.na(`gene-score`), 'Others', `gene-score`)) %>%
             left_join(GO_neuronal, by='ID') %>% left_join(clusterings, by='ID') %>%
             mutate(Neuronal=ifelse(is.na(Neuronal), 0, Neuronal)) %>%
             mutate(gene.score=ifelse(`gene-score`=='Others' & Neuronal==1, 'Neuronal', `gene-score`), 
                    significant=padj<0.05 & !is.na(padj))

# Add gene symbol
getinfo = c('ensembl_gene_id','external_gene_id')
mart = useMart(biomart='ENSEMBL_MART_ENSEMBL', dataset='hsapiens_gene_ensembl',
               host='feb2014.archive.ensembl.org') ## Gencode v19
gene_names = getBM(attributes=getinfo, filters=c('ensembl_gene_id'), values=genes_info$ID, mart=mart)

genes_info = genes_info %>% left_join(gene_names, by=c('ID'='ensembl_gene_id'))


clustering_selected = 'DynamicHybrid'
genes_info$Module = genes_info[,clustering_selected]

dataset = read.csv(paste0('./../Data/dataset_', clustering_selected, '.csv'))
dataset$Module = dataset[,clustering_selected]

load('./../Data/GSEA.RData')
GSEA_SFARI = enrichment_SFARI

load('./../Data/ORA.RData')
ORA_SFARI = enrichment_SFARI


rm(DE_info, GO_annotations, clusterings, getinfo, mart, efit, enrichment_DGN, enrichment_DO, enrichment_GO,
   enrichment_KEGG, enrichment_Reactome, enrichment_SFARI)

Selecting Top Modules

We have results both from GSEA and ORA to measure the enrichment of SFARI Genes in each module, and they both agree with each other relatively well

SFARI_genes_by_module = c()

for(module in names(GSEA_SFARI)){
  
  GSEA_info = GSEA_SFARI[[module]] %>% dplyr::select(ID, pvalue, p.adjust, NES) %>%
              mutate(pvalue = ifelse(NES>0, pvalue, 1-pvalue), 
                     p.adjust = ifelse(NES>0, p.adjust, 1)) %>%
              dplyr::rename('GSEA_pval' = pvalue, 'GSEA_padj'= p.adjust)
  
  ORA_info = ORA_SFARI[[module]] %>% dplyr::select(ID, pvalue, p.adjust, qvalue, GeneRatio, Count) %>%
             dplyr::rename('ORA_pval' = pvalue, 'ORA_padj' = p.adjust)
  
  module_info = GSEA_info %>% full_join(ORA_info, by = 'ID') %>% add_column(.before = 'ID', Module = module)
  
  SFARI_genes_by_module = rbind(SFARI_genes_by_module, module_info)
}

SFARI_genes_by_module = SFARI_genes_by_module %>% 
                        left_join(dataset %>% dplyr::select(Module, MTcor) %>% 
                                  group_by(Module,MTcor) %>% tally %>% ungroup, by = 'Module') %>%
                        mutate(ORA_pval = ifelse(is.na(ORA_pval), 1, ORA_pval),
                               ORA_padj = ifelse(is.na(ORA_padj), 1, ORA_padj))

plot_data = SFARI_genes_by_module %>% filter(ID=='SFARI')

ggplotly(plot_data %>% ggplot(aes(1-GSEA_pval, 1-ORA_pval, size = n)) + 
         geom_point(color = plot_data$Module, alpha = .6, aes(id=Module)) + 
         geom_smooth(se=FALSE, color = '#CCCCCC') + 
         xlab('GSEA Enrichment') + ylab('ORA Enrichment') + coord_fixed() +
         ggtitle(paste0('Corr = ', round(cor(plot_data$GSEA_pval, plot_data$ORA_pval),2))) +
         theme_minimal() + theme(legend.position = 'none'))

To determine which modules have a statistically significant enrichment in SFARI Genes we can use the adjusted p-values. We used the Bonferroni correction for this.

GSEA identifies 24/95 as significant. This doesn’t make sense

ggplotly(plot_data %>% ggplot(aes(GSEA_padj, ORA_padj, size = n)) + 
         geom_point(color = plot_data$Module, alpha = .7, aes(id=Module)) + 
         geom_smooth(se=FALSE, color = '#CCCCCC') + 
         geom_hline(yintercept = 0.01, color = 'gray', linetype = 'dashed') +
         geom_vline(xintercept = 0.01, color = 'gray', linetype = 'dashed') +
         xlab('GSEA adjusted p-value') + ylab('ORA adjusted p-value') + 
         scale_x_log10(limits = c(min(plot_data$GSEA_padj, plot_data$ORA_padj),1.2)) + 
         scale_y_log10(limits = c(min(plot_data$GSEA_padj, plot_data$ORA_padj),1.2)) +
         ggtitle(paste0('Corr = ',round(cor(plot_data$GSEA_padj, plot_data$ORA_padj),2))) + coord_fixed() +
         theme_minimal() + theme(legend.position = 'none'))

plot_data = plot_data %>% mutate(GSEA_sig = GSEA_padj<0.01, ORA_sig = ORA_padj<0.01) %>%
            apply_labels(GSEA_sig = 'GSEA significant enrichment',
                         ORA_sig = 'ORA significant enrichment')

cro(plot_data$GSEA_sig, list(plot_data$ORA_sig, total()))

	ORA significant enrichment		#Total
	FALSE	TRUE
GSEA significant enrichment
FALSE	70	1	71
TRUE	23	1	24
#Total cases	93	2	95

The ‘over-enrichment’ in SFARI Modules in GSEA could be because SFARI Genes have in general higher Module Memberships than the other genes, which would make them cluster at the beginning of the list constantly and would bias the enrichment analysis.

Looking at the plot below, we can see that there is not a uniform distribution of SFARI genes across all quantiles of the Module Membership values, but they instead seem to cluster around Module Membership values with high magnitudes (both positive and negative), so I don’t think the GSEA results for the SFARI genes are valid.

Because of this, I’m going to use the enrichment from the ORA to study the SFARI Genes

quant_data = dataset %>% dplyr::select(ID, contains('MM.')) %>% 
             left_join(genes_info %>% dplyr::select(ID, gene.score), by = 'ID') %>% dplyr::select(-ID) %>%
             melt %>% mutate(quant = cut(value, breaks = quantile(value, probs = seq(0,1,0.05)) %>% 
                                     as.vector, labels = FALSE)) %>%
             group_by(gene.score, quant) %>% tally %>% ungroup %>% ungroup
  
quant_data = quant_data %>% group_by(quant) %>% summarise(N = sum(n)) %>% ungroup %>% 
            left_join(quant_data, by = 'quant') %>% dplyr::select(quant, gene.score, n, N) %>% 
            mutate(p = round(100*n/N,2)) %>% filter(!is.na(quant)) %>%
            mutate(gene.score = factor(gene.score, levels = rev(c('1','2','3','Neuronal','Others'))))

ggplotly(quant_data %>% filter(!gene.score %in% c('Neuronal','Others')) %>% 
         ggplot(aes(quant, p, fill = gene.score)) + geom_bar(stat='identity') + 
         xlab('Module Membership Quantiles') + ylab('% of SFARI Genes in Quantile') +
         ggtitle('Percentage of Genes labelled as SFARI in each Quantile') +
         scale_fill_manual(values = SFARI_colour_hue(r=rev(c(1:3)))) + 
         theme_minimal() + theme(legend.position = 'none'))

rm(quant_data)

Selecting modules with an adjusted p-value below 0.01 using the ORA

ggplotly(plot_data %>% ggplot(aes(MTcor, ORA_padj, size=n)) + 
         geom_point(color=plot_data$Module, alpha=0.5, aes(id=Module)) +
         geom_hline(yintercept = 0.01, color = 'gray', linetype = 'dotted') + 
         xlab('Module-Diagnosis Correlation') + ylab('Corrected p-values') + scale_y_log10() +
         ggtitle('Modules Significantly Enriched in SFARI Genes') +
         theme_minimal() + theme(legend.position = 'none'))

top_modules = plot_data %>% arrange(desc(ORA_padj)) %>% filter(ORA_padj<0.01) %>% pull(Module) %>% as.character

plot_data %>% filter(Module %in% top_modules) %>% arrange(ORA_pval) %>%
              dplyr::select(Module, MTcor, ORA_pval, ORA_padj, qvalue, GeneRatio, Count) %>%
              rename( ORA_pval = 'p-value', ORA_padj = 'Adjusted p-value') %>%
              kable %>% kable_styling(full_width = F)

Module	MTcor	p-value	Adjusted p-value	qvalue	GeneRatio	Count
#A789FF	0.6112554	0.0000307	0.0001537	0.0000324	40/407	40
#FF64B1	0.5043320	0.0013666	0.0068330	0.0028771	21/205	21

pca = datExpr %>% prcomp

plot_data = data.frame('ID'=rownames(datExpr), 'PC1' = pca$x[,1], 'PC2' = pca$x[,2]) %>%
            left_join(dataset, by='ID') %>% 
            left_join(genes_info %>% dplyr::select(ID, external_gene_id), by='ID') %>%
            dplyr::select(ID, external_gene_id, PC1, PC2, Module, gene.score) %>%
            mutate(ImportantModules = ifelse(Module %in% top_modules, as.character(Module), 'Others')) %>%
            mutate(color = ifelse(ImportantModules=='Others','gray',ImportantModules),
                   alpha = ifelse(ImportantModules=='Others', 0.1, 0.6),
                   gene_id = paste0(ID, ' (', external_gene_id, ')')) %>%
            apply_labels(ImportantModules = 'Top Modules')

plot_data %>% ggplot(aes(PC1, PC2, color=ImportantModules)) + 
              geom_point(alpha=plot_data$alpha, color=plot_data$color, aes(ID=gene_id)) + theme_minimal() +
              xlab(paste0('PC1 (',round(100*summary(pca)$importance[2,1],2),'%)')) +
              ylab(paste0('PC2 (',round(100*summary(pca)$importance[2,2],2),'%)')) +
              ggtitle('Genes belonging to the Modules with the highest SFARI Genes Enrichment')

rm(pca)

Identifying representative genes for each Module

Following the WGCNA pipeline, selecting the genes with the highest Module Membership and Gene Significance

Top 20 genes for each module

Ordered by \(\frac{MM+|GS|}{2}\)

There aren’t that many SFARI genes in the top genes of the modules

create_table = function(module){
  top_genes = dataset %>% left_join(genes_info %>% dplyr::select(ID, external_gene_id), by='ID') %>% 
              dplyr::select(ID, external_gene_id, paste0('MM.',gsub('#','',module)), GS, gene.score) %>%
              filter(dataset$Module==module) %>% dplyr::rename('MM' = paste0('MM.',gsub('#','',module))) %>% 
              mutate(Relevance = (MM+abs(GS))/2) %>% arrange(by=-Relevance) %>% top_n(20) %>%
              dplyr::rename('Gene Symbol' = external_gene_id, 'SFARI Score' = gene.score)
  return(top_genes)
}

top_genes = list()
for(i in 1:length(top_modules)) top_genes[[i]] = create_table(top_modules[i])

kable(top_genes[[1]] %>% dplyr::select(-ID), caption=paste0('Top 20 genes for Module ', top_modules[1], 
      '  (MTcor = ', round(dataset$MTcor[dataset$Module == top_modules[1]][1],2),')')) %>% 
      kable_styling(full_width = F)

Top 20 genes for Module #FF64B1 (MTcor = 0.5)
Gene Symbol	MM	GS	SFARI Score	Relevance
TBC1D23	0.7884941	0.6872194	Others	0.7378567
CBX5	0.8148188	0.6007890	Others	0.7078039
VPS41	0.7832445	0.5614859	Others	0.6723652
FBXW7	0.8611394	0.4512806	Others	0.6562100
CREBBP	0.7233588	0.5841394	1	0.6537491
PPP2R5C	0.7357587	0.5163063	Others	0.6260325
ZRANB2	0.8757220	0.3735327	Others	0.6246274
FAM13C	0.7390301	0.5089977	Others	0.6240139
TPM3	0.7293832	0.5179388	Others	0.6236610
RORA	0.6819642	0.5602203	Others	0.6210923
STAU1	0.6783867	0.5620162	Others	0.6202015
MEF2A	0.6802534	0.5299537	Others	0.6051035
RRBP1	0.6733003	0.5316807	Others	0.6024905
NFAT5	0.7199273	0.4840942	Others	0.6020108
TAF13	0.7462561	0.4537740	Others	0.6000150
TCEA1	0.6666325	0.5320542	Others	0.5993434
LEO1	0.7175695	0.4609003	2	0.5892349
GABRA3	0.6939019	0.4632479	Others	0.5785749
ARAP2	0.6369266	0.5001257	Others	0.5685262
OSBPL8	0.7292083	0.3926742	Others	0.5609413

kable(top_genes[[2]] %>% dplyr::select(-ID), caption=paste0('Top 20 genes for Module ', top_modules[2], 
      '  (MTcor = ', round(dataset$MTcor[dataset$Module == top_modules[2]][1],2),')')) %>% 
      kable_styling(full_width = F)

Top 20 genes for Module #A789FF (MTcor = 0.61)
Gene Symbol	MM	GS	SFARI Score	Relevance
CPNE3	0.7163190	0.7410235	Others	0.7286712
JADE3	0.6615278	0.7190986	Others	0.6903132
LRRN1	0.7265126	0.6513188	Others	0.6889157
TMSB15A	0.6928634	0.6842224	Others	0.6885429
NLGN4X	0.8348708	0.5354803	2	0.6851755
DCAF12	0.7808185	0.5685263	Others	0.6746724
FAM172A	0.6926313	0.6366515	Others	0.6646414
BCHE	0.7214337	0.6063030	Others	0.6638684
CNR1	0.7731004	0.5261326	2	0.6496165
PHF21A	0.6551561	0.6348232	1	0.6449896
SMAD3	0.6573250	0.6094305	Others	0.6333778
ZMIZ1	0.6273598	0.6240369	Others	0.6256983
NFIB	0.7610366	0.4874557	3	0.6242462
PDGFRA	0.7463291	0.4996192	Others	0.6229741
NUDCD1	0.7392538	0.4902374	Others	0.6147456
SLC16A2	0.6005639	0.6210378	Others	0.6108008
E2F5	0.6781056	0.5405444	Others	0.6093250
WSCD1	0.7097529	0.5024290	Others	0.6060910
RRP15	0.5194008	0.6792841	Others	0.5993424
MN1	0.7073455	0.4857411	Others	0.5965433

rm(create_table, i)

pca = datExpr %>% prcomp

ids = c()
for(tg in top_genes) ids = c(ids, tg$ID)

plot_data = data.frame('ID'=rownames(datExpr), 'PC1' = pca$x[,1], 'PC2' = pca$x[,2]) %>%
            left_join(dataset, by='ID') %>% dplyr::select(ID, PC1, PC2, Module, gene.score) %>%
            mutate(color = ifelse(Module %in% top_modules, as.character(Module), 'gray')) %>%
            mutate(alpha = ifelse(color %in% top_modules & ID %in% ids, 1, 0.1))

plot_data %>% ggplot(aes(PC1, PC2)) + geom_point(alpha=plot_data$alpha, color=plot_data$color) + 
              xlab(paste0('PC1 (',round(100*summary(pca)$importance[2,1],2),'%)')) +
              ylab(paste0('PC2 (',round(100*summary(pca)$importance[2,2],2),'%)')) +
              theme_minimal() + ggtitle('Most relevant genes for top Modules')

rm(ids, pca, tg, plot_data)

Enrichment Analysis

Using the package clusterProfiler. Performing Gene Set Enrichment Analysis (GSEA) and Over Representation Analysis (ORA) using the following datasets:

Gene Ontology
Disease Ontology
Disease Gene Network
KEGG
REACTOME

# GSEA
load('./../Data/GSEA.RData')

# Rename lists
GSEA_GO = enrichment_GO
GSEA_DGN = enrichment_DGN
GSEA_DO = enrichment_DO
GSEA_KEGG = enrichment_KEGG
GSEA_Reactome = enrichment_Reactome
GSEA_SFARI = enrichment_SFARI


# ORA
load('./../Data/ORA.RData')

# Rename lists
ORA_GO = enrichment_GO
ORA_DGN = enrichment_DGN
ORA_DO = enrichment_DO
ORA_KEGG = enrichment_KEGG
ORA_Reactome = enrichment_Reactome
ORA_SFARI = enrichment_SFARI

rm(enrichment_GO, enrichment_DO, enrichment_DGN, enrichment_KEGG, enrichment_Reactome)

compare_methods = function(GSEA_list, ORA_list){
  
  for(top_module in top_modules){
  
    cat(paste0('  \n  \nEnrichments for Module ', top_module, ' (MTcor=',
               round(dataset$MTcor[dataset$Module==top_module][1],2), '):  \n  \n'))
    
    GSEA = GSEA_list[[top_module]]
    ORA = ORA_list[[top_module]]
    
    cat(paste0('GSEA has ', nrow(GSEA), ' enriched terms  \n'))
    cat(paste0('ORA has  ', nrow(ORA), ' enriched terms  \n'))
    cat(paste0(sum(ORA$ID %in% GSEA$ID), ' terms are enriched in both methods  \n  \n'))

    plot_data = GSEA %>% mutate(pval_GSEA = p.adjust) %>% dplyr::select(ID, Description, NES, pval_GSEA) %>%
                inner_join(ORA %>% mutate(pval_ORA = p.adjust) %>% 
                           dplyr::select(ID, pval_ORA, GeneRatio, qvalue), by = 'ID') 
    
    if(nrow(plot_data)>0){
      print(plot_data %>% mutate(pval_mean = pval_ORA + pval_GSEA) %>% 
                          arrange(pval_mean) %>% dplyr::select(-pval_mean) %>% 
            kable %>% kable_styling(full_width = F))
    }
  } 
}


plot_results = function(GSEA_list, ORA_list){
  
  l = htmltools::tagList()

  for(i in 1:length(top_modules)){
    
    GSEA = GSEA_list[[top_modules[i]]]
    ORA = ORA_list[[top_modules[i]]]
    
    plot_data = GSEA %>% mutate(pval_GSEA = p.adjust) %>% dplyr::select(ID, Description, NES, pval_GSEA) %>%
                inner_join(ORA %>% mutate(pval_ORA = p.adjust) %>% dplyr::select(ID, pval_ORA), by = 'ID')
    
    if(nrow(plot_data)>5){
      min_val = min(min(plot_data$pval_GSEA), min(plot_data$pval_ORA))
      max_val = max(max(max(plot_data$pval_GSEA), max(plot_data$pval_ORA)),0.05)
      ggp = ggplotly(plot_data %>% ggplot(aes(pval_GSEA, pval_ORA, color = NES)) + 
                     geom_point(aes(id = Description)) + 
                     geom_vline(xintercept = 0.05, color = 'gray', linetype = 'dotted') + 
                     geom_hline(yintercept = 0.05, color = 'gray', linetype = 'dotted') + 
                     ggtitle(paste0('Enriched terms in common for Module ', top_modules[i])) +
                     scale_x_continuous(limits = c(min_val, max_val)) + 
                     scale_y_continuous(limits = c(min_val, max_val)) + 
                     xlab('Corrected p-value for GSEA') + ylab('Corrected p-value for ORA') +
                     scale_colour_viridis(direction = -1) + theme_minimal() + coord_fixed())
      l[[i]] = ggp
    }
  }
  
  return(l)
}